Labeling Method for Road Traffic Lights, Autonomous Driving Computing Platform and Storage Medium

Abstract

The present application relates to the technical field of object annotation, and discloses a method for annotating a road traffic light, an autonomous driving computing platform, and a storage medium. The method includes: acquiring a road image and a corresponding initial point cloud; identifying a traffic light area in the road image; determining a first target point cloud area corresponding to the traffic light area in the initial point cloud; clustering the first target point cloud area to obtain at least one first point cloud cluster; and annotating the point cloud cluster corresponding to the traffic light in the at least one first point cloud cluster. By the above method, the accuracy of traffic light annotation can be improved, and incorrect annotation can be reduced.

Description

Technical Field

[0001] This application relates to the technical field of object annotation, and particularly to a method for annotating road traffic lights, an autonomous driving computing platform, and a storage medium. Background Art

[0002] In the fields of autonomous driving, high-precision maps, etc., crowdsourcing-based acquisition and mapping with high update frequency, small computational load, and fast transmission are becoming increasingly popular. The on-vehicle sensors used for map acquisition generally mainly include LiDAR, cameras, RTK GPS, etc. Both LiDAR and cameras can sense the vehicle's surrounding environment and provide positioning functions. RTK GPS can provide millimeter-level positioning accuracy in the case of good signals. In addition to positioning information, the traffic element information such as roads, traffic signs, obstacles, and pedestrians obtained by sensing will be used to control the vehicle's steering and speed.

[0003] The timely and accurate recognition of traffic lights directly affects the safety and reliability of autonomous driving vehicles. In related technologies, there are many false detection and missed detection phenomena in the process of traffic light recognition and annotation. For example, the points on tree leaves can also be projected into the bounding box in the image, resulting in false detection. And in order to filter out these false detected objects, it is possible to delete the correct traffic light point cloud, resulting in missed detection. Summary of the Invention

[0004] The main technical problem to be solved by this application is to provide a method for annotating road traffic lights, an autonomous driving computing platform, and a storage medium, which can improve the accuracy of traffic light annotation and reduce incorrect annotation.

[0005] To solve the above technical problem, a technical solution adopted by this application is: to provide a method for annotating road traffic lights, the method includes: obtaining a road image and the corresponding initial point cloud; identifying the traffic light area in the road image; determining the first target point cloud area corresponding to the traffic light area in the initial point cloud; clustering the first target point cloud area to obtain at least one first point cloud cluster; annotating the point cloud clusters corresponding to the traffic lights in the at least one first point cloud cluster.

[0006] Among them, annotating the point cloud clusters corresponding to the traffic lights in the at least one first point cloud cluster includes: performing plane segmentation on the target point cloud clusters corresponding to the traffic lights in the at least one first point cloud cluster to obtain the second target point cloud corresponding to the same plane; annotating the second target point cloud.

[0007] Among them, performing plane segmentation on the target point cloud cluster corresponding to the traffic light in at least one first point cloud cluster to obtain the second target point cloud corresponding to the same plane includes: establishing an initial plane based on the target point cloud cluster; wherein, the angle between the initial plane and the Z-axis of the world coordinate system is less than a preset angle; dividing the target points in the target point cloud cluster that meet the preset conditions into the initial plane; if the number of target points in the initial plane meets the preset ratio, taking the target points in the initial plane as the second target point cloud; wherein, the preset condition is that the perpendicular distance from the target point to the initial plane is less than the first distance.

[0008] Among them, annotating the point cloud cluster corresponding to the traffic light in at least one first point cloud cluster includes: performing line segmentation on the target point cloud cluster corresponding to the traffic light in at least one first point cloud cluster to obtain the second target point cloud corresponding to the same line; annotating the second target point cloud.

[0009] Among them, performing line segmentation on the target point cloud cluster corresponding to the traffic light in at least one first point cloud cluster to obtain the second target point cloud corresponding to the same line includes: establishing an initial line based on the target point cloud cluster; if the number of target points in the initial line meets the preset ratio, taking the target points in the initial line as the second target point cloud.

[0010] Among them, after clustering the first target point cloud region to obtain at least one first point cloud cluster, it includes: obtaining at least one first point cloud cluster corresponding to at least two frames of road images; clustering the point cloud clusters corresponding to the traffic light in all at least one first point cloud clusters to obtain at least one second point cloud cluster; annotating the point cloud clusters corresponding to the traffic light in at least one first point cloud cluster includes: taking the second point cloud cluster as the first point cloud cluster and annotating the point cloud clusters corresponding to the traffic light in at least one second point cloud cluster.

[0011] Among them, annotating the point cloud clusters corresponding to the traffic light in at least one second point cloud cluster includes: performing plane segmentation and / or line segmentation on the target point cloud clusters corresponding to the traffic light in at least one second point cloud cluster to obtain the third target point cloud corresponding to the same plane or the same line; annotating the third target point cloud.

[0012] Among them, annotating the third target point cloud includes: determining the center point of the third target point cloud; annotating the center point.

[0013] To solve the above technical problems, another technical solution adopted by this application is: providing an autonomous driving computing platform, which includes a processor and a memory coupled to the processor, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method provided by the above technical solution.

[0014] To solve the above technical problems, another technical solution adopted by this application is: to provide a computer-readable storage medium for storing a computer program, which, when executed by a processor, is used to implement the method provided by the above technical solution.

[0015] The beneficial effects of the embodiments of this application are as follows: Different from the prior art, the method for annotating a road traffic light provided by this application includes: obtaining a road image and the corresponding initial point cloud; identifying the traffic light area in the road image; determining the first target point cloud area in the initial point cloud corresponding to the traffic light area; clustering the first target point cloud area to obtain at least one first point cloud cluster; and annotating the point cloud cluster corresponding to the traffic light in the at least one first point cloud cluster. By the above method, after determining the first target point cloud area in the initial point cloud corresponding to the traffic light area, clustering the first target point cloud area to remove the points that are not traffic lights, and then annotating the points belonging to the traffic lights, the accuracy of traffic light annotation can be improved and incorrect annotation can be reduced. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0017] Figure 1 is a schematic flowchart of an embodiment of the method for annotating a road traffic light provided by this application;

[0018] Figure 2 is a schematic diagram of an application scenario of the method for annotating a road traffic light provided by this application;

[0019] Figure 3 is a schematic flowchart of another embodiment of the method for annotating a road traffic light provided by this application;

[0020] Figure 4 is a schematic flowchart of an embodiment of step 35 provided by this application;

[0021] Figure 5 is a schematic flowchart of another embodiment of the method for annotating a road traffic light provided by this application;

[0022] Figure 6 is a schematic flowchart of an embodiment of step 55 provided by this application;

[0023] Figure 7 is a schematic flowchart of another embodiment of the method for annotating a road traffic light provided by this application;

[0024] Figure 8 It is a schematic flowchart of an embodiment of step 77 provided by this application;

[0025] Figure 9 It is a schematic flowchart of an embodiment of step 772 provided by this application;

[0026] Figure 10 It is a schematic diagram of an application scenario of related art provided by this application;

[0027] Figure 11 It is a schematic diagram of an application scenario of a method for annotating road traffic lights provided by this application;

[0028] Figure 12 It is a schematic diagram of an application scenario of related art provided by this application;

[0029] Figure 13 It is a schematic diagram of an application scenario of a method for annotating road traffic lights provided by this application;

[0030] Figure 14 It is a schematic flowchart of an embodiment of an autonomous driving computing platform provided by this application;

[0031] Figure 15 It is a schematic structural diagram of an embodiment of a computer-readable storage medium provided by this application. Detailed implementation manners

[0032] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. It can be understood that the specific embodiments described herein are only used to explain this application, rather than limiting this application. Additionally, it should be noted that for the sake of description, only parts related to this application rather than all structures are shown in the accompanying drawings. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0033] Referring to "embodiment" in this context means that the specific features, structures, or characteristics described in conjunction with the embodiment may be included in at least one embodiment of this application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0034] During the traffic light annotation process, there are usually phenomena of missed annotation or misannotation. For example, points on tree leaves can also be projected into the bounding boxes in the image, resulting in false detections. In order to filter out these false detected objects, it is possible to delete the correct traffic light point clouds, resulting in missed detections. Based on this, the present application uses clustering to distinguish traffic lights from non-traffic lights for annotation. For specific reference, please refer to the following embodiments.

[0035] Refer to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of the method for annotating road traffic lights provided by the present application. The method includes:

[0036] Step 11: Obtain a road image and the corresponding initial point cloud.

[0037] Among them, a road image can be collected using an image sensor, and a point cloud can be collected using a radar sensor. For example, the image sensor and the radar sensor are installed on a movable device. Among them, the device can be an automatic moving device, such as a robot, an autonomous vehicle, etc.

[0038] In some embodiments, the image sensor can be a camera. The radar sensor can be a lidar sensor, such as a mechanical lidar.

[0039] In a corresponding scenario, an autonomous vehicle is driving on the road, and a road image is obtained through an image sensor installed on the autonomous vehicle and the corresponding initial point cloud is obtained using a radar sensor.

[0040] Step 12: Identify the traffic light area in the road image.

[0041] During the driving process of the autonomous vehicle, the obtained road image usually includes roads, traffic lights, trees, street lights, buildings, etc. around the road. Therefore, through image recognition technology, the traffic light area in the road image is identified.

[0042] Among them, the traffic light can be a traffic light, a traffic sign, etc.

[0043] After identifying the traffic light area in the road image, the traffic light area can be framed using a rectangular box.

[0044] Step 13: Determine the first target point cloud area corresponding to the traffic light area in the initial point cloud.

[0045] In some embodiments, since the initial point cloud has three-dimensional information and the road image is two-dimensional, it is necessary to project the initial point cloud onto the plane where the road image is located, and the point cloud that falls within the traffic light area is used as the first target point cloud, and then the area where the first target point cloud is located is used as the first target point cloud area.

[0046] As shown in Figure 2 , the traffic light regions A and B are identified in the road image through step 12. Meanwhile, the initial point cloud is projected onto the road image, and corresponding point clouds will exist in the traffic light regions A and B respectively. Furthermore, the region of the initial point cloud where the point cloud corresponding to the traffic light region A is located is determined as the first target point cloud region a, and the region of the initial point cloud where the point cloud corresponding to the traffic light region B is located is determined as the first target point cloud region b.

[0047] Step 14: Cluster the first target point cloud region to obtain at least one first point cloud cluster.

[0048] In some embodiments, due to road environmental factors, there is a situation where at least part of the traffic light is blocked. For example, when collecting the road image and the initial point cloud, the traffic light is blocked by leaves, so there are some points belonging to the leaves in the point cloud projected onto the traffic light region. Therefore, false detection of the traffic light is likely to occur. For example, taking the points belonging to the leaves as the traffic light makes the identified position of the traffic light inconsistent with the actual position of the traffic light, affecting subsequent map construction and vehicle driving based on the traffic light.

[0049] Based on this, step 14 is executed to cluster the first target point cloud region to obtain at least one first point cloud cluster.

[0050] In some embodiments, the first target point cloud region can be clustered using the Euclidean clustering, K-Means clustering algorithm, etc. to obtain at least one first point cloud cluster.

[0051] That is, through the clustering algorithm, the points in the first target point cloud region can be divided, and further divided into the first point cloud cluster belonging to the traffic light and the first point cloud cluster belonging to non-traffic lights.

[0052] Step 15: Label the point cloud cluster corresponding to the traffic light in at least one first point cloud cluster.

[0053] In step 15, the center point in the point cloud cluster corresponding to the traffic light can be determined and labeled.

[0054] In some embodiments, since the coordinates of the points in the initial point cloud are based on the radar sensor, after determining the point cloud cluster corresponding to the traffic light, the coordinates of the points in the point cloud cluster corresponding to the traffic light are converted in the coordinate system to the coordinates corresponding to the points in the point cloud cluster in the world coordinate system. Then the center point is labeled.

[0055] In some embodiments, as shown in Figure 2 , because Figure 2Traffic light regions A and B appear. Then, steps 14 - 15 can be respectively executed for each traffic light region to label the point cloud clusters of the corresponding traffic lights in both traffic light region A and traffic light region B. That is, in Figure 2 two traffic lights will be labeled.

[0056] In this embodiment, by acquiring a road image and the corresponding initial point cloud; identifying the traffic light regions in the road image; determining the first target point cloud region corresponding to the traffic light regions in the initial point cloud; clustering the first target point cloud region to obtain at least one first point cloud cluster; and labeling the point cloud clusters corresponding to the traffic lights in at least one first point cloud cluster, after determining the first target point cloud region corresponding to the traffic light regions in the initial point cloud, clustering the first target point cloud region to remove the points that are not traffic lights, and then labeling the points belonging to the traffic lights, the accuracy of traffic light labeling can be improved and mislabeling can be reduced.

[0057] Refer to Figure 3 , Figure 3 which is a schematic flowchart of another embodiment of the method for labeling road traffic lights provided by this application. The method includes:

[0058] Step 31: Acquire a road image and the corresponding initial point cloud.

[0059] Step 32: Identify the traffic light regions in the road image.

[0060] Step 33: Determine the first target point cloud region corresponding to the traffic light regions in the initial point cloud.

[0061] Step 34: Cluster the first target point cloud region to obtain at least one first point cloud cluster.

[0062] Steps 31 - 34 have the same or similar technical solutions as any of the above embodiments and will not be elaborated here.

[0063] In some embodiments, since the road image and the initial point cloud are acquired using different sensors, there is a difference in the acquisition time. Therefore, it is necessary to align the timestamps of the road image and the initial point cloud. That is, obtain the road image and the initial point cloud with the same timestamp.

[0064] Furthermore, if the initial point cloud is obtained during the driving of an autonomous vehicle, motion compensation needs to be performed on the initial point cloud. For a mechanical lidar, such as a 64-line lidar, its general principle is to stack 64 lasers vertically together and rotate the entire unit many times per second. There is physical rotation in the emission system and the reception system. That is, by continuously rotating the emitter, the laser points are turned into lines, and multiple laser emitters are arranged vertically to form a plane, achieving the purpose of 3D scanning and receiving information. As described above, it is not difficult to understand that during the movement of an autonomous vehicle, the points received by the lidar in each frame are not at the same timestamp. Because the autonomous vehicle is moving and the lidar itself is also rotating.

[0065] Based on the precise positioning information of the autonomous vehicle and the rotation motion law of the lidar, each point can be conveniently converted to the same timestamp.

[0066] Step 35: Perform plane segmentation on the target point cloud cluster corresponding to the traffic light in at least one first point cloud cluster to obtain the second target point cloud corresponding to the same plane.

[0067] In some embodiments, since traffic lights are usually vertically installed on the ground, the point cloud corresponding to the traffic light is approximately on the same plane. For example, this plane can be perpendicular to the ground in the world coordinate system. Therefore, the target point cloud cluster can be further filtered by plane segmentation to obtain the second target point cloud corresponding to the same plane. The second target point cloud at this time can be considered as the point cloud corresponding to the traffic light.

[0068] In some embodiments, referring to Figure 4 , step 35 can be the following process:

[0069] Step 351: Establish an initial plane based on the target point cloud cluster; wherein, the angle between the initial plane and the Z-axis of the world coordinate system is less than a preset angle.

[0070] In some embodiments, since there are multiple points in the target point cloud cluster, an initial plane can be fitted based on these points. The initial plane should meet a condition threshold. For example, the angle between the initial plane and the Z-axis of the world coordinate system is less than a preset angle. The preset angle can be 10 degrees, 15 degrees. Among them, the Z-axis is the axis perpendicular to the ground in the world coordinate system.

[0071] When the initial plane does not meet the condition threshold, a new initial plane should be fitted based on these points again.

[0072] Step 352: Divide the target points in the target point cloud cluster that meet the preset conditions into the initial plane.

[0073] Among them, the preset condition is that the vertical distance from the target point to the initial plane is less than the first distance. For example, the first distance can be 10 cm, 5 cm, etc.

[0074] For example, if the vertical distance from the target point A in the target point cloud cluster to the initial plane is less than the first distance, then the target point A is divided into the initial plane. If the vertical distance from the target point B in the target point cloud cluster to the initial plane is greater than the first distance, then the target point B is discarded.

[0075] Step 353: If the number of target points in the initial plane meets the preset ratio, then the target points in the initial plane are used as the second target point cloud.

[0076] Since the target point cloud cluster is the point cloud determined to belong to the traffic light after clustering, the confidence level is relatively high. Therefore, when the ratio of the number of target points divided into the initial plane to the point data in the target point cloud cluster meets the preset ratio, the target points in the initial plane are used as the second target point cloud.

[0077] Among them, the preset ratio can be eighty percent, eighty-five percent or ninety percent.

[0078] Step 36: Label the second target point cloud.

[0079] In this embodiment, after determining the first target point cloud region corresponding to the traffic light region in the initial point cloud, clustering is performed on the first target point cloud region to remove the points that are not traffic lights, and then the clustered point cloud is divided by means of plane segmentation to further remove the points that are not traffic lights, and then the points belonging to the traffic light after plane segmentation are labeled, which can improve the accuracy of traffic light labeling and reduce mislabeling.

[0080] Refer to Figure 5 , Figure 5 is a schematic flowchart of another embodiment of the method for labeling road traffic lights provided by this application. The method includes:

[0081] Step 51: Obtain a road image and the corresponding initial point cloud.

[0082] Step 52: Identify the traffic light region in the road image.

[0083] Step 53: Determine the first target point cloud region corresponding to the traffic light region in the initial point cloud.

[0084] Step 54: Cluster the first target point cloud region to obtain at least one first point cloud cluster.

[0085] Steps 51 - 54 have the same or similar technical solutions as any of the above embodiments, and will not be elaborated here.

[0086] Step 55: Perform linear segmentation on the target point cloud clusters corresponding to traffic lights in at least one first point cloud cluster to obtain second target point clouds corresponding to the same straight line.

[0087] In some embodiments, due to the positional relationship between the lidar and the traffic lights, the point clouds of the traffic lights collected are not sufficient to form a plane. However, due to the working characteristics of the lidar, straight lines will be formed. Therefore, linear segmentation can be used to obtain second target point clouds corresponding to the same straight line.

[0088] In some embodiments, referring to Figure 6 , Step 55 may be the following process:

[0089] Step 551: Establish an initial straight line based on the target point cloud cluster.

[0090] In some embodiments, since there are multiple points in the target point cloud cluster, an initial straight line can be fitted based on these points.

[0091] Step 552: If the number of target points in the initial straight line meets a preset ratio, take the target points in the initial straight line as the second target point cloud.

[0092] Since the target point cloud cluster is the point cloud of the traffic light determined after clustering, the confidence level is relatively high. Therefore, when the ratio of the number of target points divided into the initial straight line to the point data in the target point cloud cluster meets the preset ratio, take the target points in the initial straight line as the second target point cloud.

[0093] Among them, the preset ratio can be eighty percent, eighty-five percent, or ninety percent.

[0094] Step 56: Label the second target point cloud.

[0095] In some embodiments, linear segmentation and plane segmentation can be performed on the target point cloud clusters corresponding to traffic lights in at least one first point cloud cluster to obtain corresponding second target point clouds. For example, first perform linear segmentation to obtain multiple straight lines, and then determine whether the multiple straight lines form a target plane.

[0096] Simultaneously perform plane segmentation to form an initial plane. Determine the difference between the target plane and the initial plane. When the difference is less than the threshold, take the target points in the target plane or the initial plane as the second target point cloud.

[0097] In this embodiment, after determining the first target point cloud region corresponding to the traffic light region in the initial point cloud, clustering is performed on the first target point cloud region to remove the points that are not traffic lights, and then the clustered point cloud is divided by means of line segmentation to further remove the points that are not traffic lights, and then the points belonging to the traffic light after plane segmentation are labeled, which can improve the accuracy of traffic light labeling and reduce mislabeling.

[0098] Refer to Figure 7 , Figure 7 which is a schematic flowchart of another embodiment of the method for labeling road traffic lights provided by this application. The method includes:

[0099] Step 71: Obtain at least two frames of road images and corresponding initial point clouds.

[0100] In some embodiments, since the autonomous mobile device travels on the road, multiple frames of road images and corresponding initial point clouds will be collected during the travel. Therefore, multiple frames of road images and corresponding initial point clouds can be processed to determine the traffic lights and label them.

[0101] Among them, the number of road images can be set according to requirements. Such as 100 frames, 200 frames. Or determined according to the acquisition time. Such as all the frames of road images and corresponding initial point clouds collected within 5 minutes.

[0102] Step 72: Identify the traffic light region in each frame of road image.

[0103] Step 73: Determine the first target point cloud region corresponding to the traffic light region in the initial point cloud of the corresponding road image.

[0104] Step 74: Cluster each first target point cloud region to obtain at least one first point cloud cluster.

[0105] Steps 72 - 74 can participate in the processing process of a single frame of road image and corresponding initial point cloud in any of the above embodiments, which will not be elaborated here.

[0106] Step 75: Obtain at least one first point cloud cluster corresponding to at least two frames of road images.

[0107] Thus, each frame of road image will correspond to at least one first point cloud cluster.

[0108] Step 76: Cluster the point cloud clusters corresponding to the traffic lights in all at least one first point cloud cluster to obtain at least one second point cloud cluster.

[0109] After step 74, there will be at least one first point cloud cluster in each frame of the initial point cloud. Then in step 76, the point cloud clusters corresponding to traffic lights in all at least one first point cloud cluster are clustered again to further classify the point cloud clusters in detail. At this time, it is equivalent to clustering all the point clouds corresponding to traffic lights in multiple frames of the initial point cloud again. In some embodiments, the point clouds processed in multiple single frames can be fused first, and then the fused point clouds of multiple frames are clustered.

[0110] Step 77: Use the second point cloud cluster as the first point cloud cluster to label the point cloud clusters corresponding to traffic lights in at least one second point cloud cluster.

[0111] In some embodiments, referring to Figure 8 , step 77 can be the following process:

[0112] Step 771: Perform plane segmentation and / or line segmentation on the target point cloud clusters corresponding to traffic lights in at least one second point cloud cluster to obtain the third target point cloud corresponding to the same plane or the same line.

[0113] For the plane segmentation and line segmentation in step 771, reference can be made to any of the above embodiments, which will not be elaborated here.

[0114] It should be noted that when performing plane segmentation and / or line segmentation, the preset conditions for multiple frames of the initial point cloud need to be set more strictly than those for a single frame of the initial point cloud. For example, when performing plane segmentation, the preset angle for multiple frames should be smaller than that for a single frame. The preset ratio of the point data for multiple frames should be greater than that for a single frame. For example, when performing line segmentation, the preset ratio of the point data for multiple frames should be greater than that for a single frame.

[0115] Step 772: Label the third target point cloud.

[0116] In some embodiments, referring to Figure 9 , step 772 can be the following process:

[0117] Step 7721: Determine the center point of the third target point cloud.

[0118] After the fourth target point cloud is determined, the center point in the fourth target point cloud is determined.

[0119] Step 7722: Label the center point.

[0120] In some embodiments, since the coordinates of the points in the initial point cloud are based on the radar sensor, after determining the point cloud cluster corresponding to the traffic light, the coordinates of the points in the point cloud cluster corresponding to the traffic light are converted to the coordinates corresponding to the points in the point cloud cluster in the world coordinate system. Then the center point is labeled.

[0121] In this embodiment, by first determining the point cloud clusters belonging to traffic lights according to a single-frame road image and the corresponding initial point cloud for multiple frames of road images and the corresponding initial point clouds, then clustering the point clouds corresponding to all frames to remove non-traffic-light points, and then using plane segmentation to divide the clustered point clouds to further remove non-traffic-light points, and then labeling the points belonging to traffic lights after plane segmentation, the accuracy of traffic light labeling can be improved and mislabeling can be reduced.

[0122] In this embodiment, processing a single-frame road image and the corresponding initial point cloud belongs to rough screening; then processing multiple frames of road images and the corresponding initial point clouds belongs to fine screening, and finally completing the labeling. Since multiple frames of road images and the corresponding initial point clouds have more traffic light information, more traffic lights can be recognized, the omission of traffic light labeling can be reduced, and the accuracy of traffic light labeling can be improved.

[0123] In an application scenario, in combination with Figures 10 - 11 it is described as follows:

[0124] It can be seen that in the technical solution of the related art Figure 10 except for the L4 traffic light, other labels are all misdetected point clouds of leaves. In Figure 11 after the method of any of the above embodiments provided by the present application, the misdetected point clouds are all filtered out.

[0125] In an application scenario, in combination with Figures 12 - 13 it is described as follows:

[0126] For example, Figure 12 and Figure 13 in Figure 12 there is actually another traffic light beside the L23 and L25 traffic lights. In Figure 13 after the method of any of the above embodiments provided by the present application, the missed traffic lights are supplemented. It can be understood that Figure 12 and Figure 13 the traffic light IDs are inconsistent, which is normal. Because the misdetections are deleted and the missed detections are supplemented, the ID numbers will be reordered.

[0127] Referring to Figure 14 Figure 14 is a schematic structural diagram of an embodiment of an autonomous driving computing platform provided by the present application. The autonomous driving computing platform 140 includes a processor 141 and a memory 142 coupled to the processor 141. The memory 142 is used to store a computer program, and the processor is used to execute the computer program to implement the following method:

[0128] ​Obtain a road image and the corresponding initial point cloud; identify the traffic light area in the road image; determine the first target point cloud area in the initial point cloud corresponding to the traffic light area; cluster the first target point cloud area to obtain at least one first point cloud cluster; label the point cloud cluster corresponding to the traffic light in at least one first point cloud cluster.

[0129] It can be understood that the processor 141 is also used to execute a computer program to implement the method provided in any of the above embodiments, which will not be elaborated here.

[0130] Refer to Figure 15 , Figure 15 FIG. is a schematic structural diagram of an embodiment of a computer-readable storage medium provided by the present application. The computer-readable storage medium 150 is used to store a computer program 151, and when the computer program 151 is executed by a processor, it is used to implement the method provided by the above technical solution.

[0131] Obtain a road image and the corresponding initial point cloud; identify the traffic light area in the road image; determine the first target point cloud area in the initial point cloud corresponding to the traffic light area; cluster the first target point cloud area to obtain at least one first point cloud cluster; label the point cloud cluster corresponding to the traffic light in at least one first point cloud cluster.

[0132] It can be understood that when the computer program 151 is executed by a processor, it is also used to implement the method provided in any of the above embodiments, which will not be elaborated here.

[0133] In several embodiments provided by the present application, it should be understood that the disclosed method and device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the circuit or unit is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0134] The unit described as a separated component may or may not be physically separated, and the component shown as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0135] In addition, each functional unit in various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0136] The above are only the embodiments of the present application, and do not thus limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made according to the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall equally be included within the patent protection scope of the present application.

Claims

1. A method for annotating a road traffic light, characterized in that, The method includes: Obtaining at least two frames of road images and corresponding initial point clouds; Identifying traffic light regions in each frame of the road images; Determining a first target point cloud region corresponding to the traffic light regions in the initial point cloud of the corresponding road image; Clustering the first target point cloud region to obtain at least one first point cloud cluster; Obtaining the at least one first point cloud cluster corresponding to at least two frames of road images; Clustering the point cloud clusters corresponding to traffic lights among all the at least one first point cloud clusters to obtain at least one second point cloud cluster; Performing plane segmentation and / or line segmentation on the target point cloud cluster corresponding to traffic lights in the at least one second point cloud cluster to obtain a third target point cloud corresponding to the same plane or the same line; wherein, when performing plane segmentation, the preset angle corresponding to multiple frames is smaller than the preset angle of a single frame; when performing plane segmentation and / or line segmentation, the preset ratio of the point data of multiple frames is larger than the preset ratio of a single frame; Labeling the third target point cloud.

2. The method according to claim 1, wherein The performing plane segmentation on the target point cloud cluster corresponding to traffic lights in the at least one second point cloud cluster to obtain a third target point cloud corresponding to the same plane includes: Establishing an initial plane based on the target point cloud cluster; wherein, the angle between the initial plane and the Z-axis of the world coordinate system is smaller than the preset angle; Dividing the target points in the target point cloud cluster that meet the preset conditions into the initial plane; If the number of target points in the initial plane meets the preset ratio, taking the target points in the initial plane as the third target point cloud; Wherein, the preset condition is that the perpendicular distance from the target point to the initial plane is smaller than the first distance.

3. The method according to claim 1, wherein The performing line segmentation on the target point cloud cluster corresponding to traffic lights in the at least one second point cloud cluster to obtain a third target point cloud corresponding to the same line: Establishing an initial line based on the target point cloud cluster; If the number of target points in the initial line meets the preset ratio, taking the target points in the initial line as the third target point cloud.

4. The method according to claim 1, wherein The labeling the third target point cloud includes: Determining the center point of the third target point cloud; Labeling the center point.

5. An autonomous driving computing system, characterized in that, The autonomous driving computing system includes a processor and a memory coupled to the processor, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method according to any one of claims 1-4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, and the computer program, when executed by a processor, is used to implement the method according to any one of claims 1-4.

Images